AI Bubble or Real Growth? Why Both Can Be True

AI can create real economic value while some investments built around it fail to earn what their owners expected. That is why “AI works” and “parts of the AI investment boom may be overextended” are not contradictory statements.

The debate becomes less useful when those statements are forced into a single yes-or-no verdict. A better question is where useful activity turns into dependable cash, and what assumptions sit between the two.

Evidence reviewed: 26 October 2026. This is an educational analysis, not a declaration that an entire asset class is a bubble or a prediction of when prices will move. The examples below are invented, not Altys company data or forecasts.

One technology, four different tests

First, does the tool solve a worthwhile problem? A customer might finish a task faster, reduce an error rate or provide a service that was previously impractical.

Second, does the customer capture a financial benefit after paying for the tool and checking its output? Time saved is not automatically money saved. The employee may still be employed, the output may need review, and the organisation may not have additional demand for the capacity released.

Third, does the supplier earn cash after providing the service and funding the assets behind it? A growing customer base can coexist with rising delivery costs and repeated capital requirements.

Fourth, what expectations are already embedded in the price of the investment? Even a successful business can disappoint an owner who paid for a much stronger outcome. This is a general valuation principle, not a conclusion about a named security.

Test	Evidence that helps	What it does not establish
Usefulness	A measured task improvement	The customer’s total financial return
Adoption	Renewing, paying customers	Supplier profitability
Business economics	Cash after costs and necessary reinvestment	An attractive purchase price
Valuation	Explicit future-cash assumptions	A reliable market-timing signal

These tests can deliver different answers at the same time. The research problem is to keep them separate.

What the recent official risk discussion says

The Bank of England’s September 2026 Financial Policy Committee record, published on 30 September, discusses AI-related valuations, earnings expectations and growing debt financing. It reports that the July adjustment did not spill into core markets, while retaining concern about sharper repricing. This is a financial-stability assessment, not a dated crash forecast.

The IMF’s April 2026 Global Financial Stability Report, Chapter 1, also examines AI financing and technological obsolescence. That is an older analytical reference, not an October observation. Its discussion distinguishes strong existing balance sheets from possible future funding pressures.

Altys interpretation: a sound investigation tests the financing and cash assumptions without assuming either that all spending is wasteful or that technological progress validates every investment.

Spending arrives before the accounting expense

Imagine a business buys computing equipment for ₹120 crore in cash at the start of a year. Assume it is immediately available for use, has no residual value and is depreciated evenly over its estimated useful life. Ignore tax, financing, impairment and every other asset.

Under a three-year estimate, annual depreciation is ₹120 crore ÷ 3 = ₹40 crore. Under six years, it is ₹120 crore ÷ 6 = ₹20 crore. If annual revenue and cash operating costs leave ₹60 crore before depreciation, the corresponding operating profits are ₹20 crore and ₹40 crore.

Invented sensitivity, not a company observation. Revenue less cash operating costs is ₹60 crore in both cases; equipment costs ₹120 crore, bought at the start and immediately in use; zero residual value, straight-line depreciation, no tax, financing or impairment.

Nothing about the longer estimate recovers the ₹120 crore cash outlay sooner. If all revenue is collected and operating costs are paid in that year, with no working-capital movement, the same simplified first-year cash arithmetic is ₹60 crore less ₹120 crore = a ₹60 crore outflow before the omitted items.

This is not evidence that a six-year estimate is wrong. Equipment can remain commercially useful after a newer product appears. Equally, an accounting life does not guarantee that customers will pay enough for its output throughout that period.

The IFRS Foundation’s IAS 16 material explains systematic depreciation over useful life and review of the estimate. This is general IFRS context, not a statement that every US or Indian company uses the same local accounting rules.

Utilisation is not the same as pricing power

A busy facility can still have difficult economics. More usage might arrive because prices fell, because promotional credits were offered or because a customer signed a contract that is expensive to serve.

Track paid usage separately from capacity, headline demand and free trials. Then examine the cash cost of delivering that usage, contract duration and renewal pricing. A model that only increases utilisation while leaving price and operating cost unchanged is making several assumptions, not one.

Replacement spending adds another distinction. Money spent to expand capacity serves a different purpose from money needed to maintain a commercially competitive service. Public disclosures may not separate those amounts cleanly. Where they do not, show alternative assumptions rather than manufacture a precise maintenance-capex figure.

Our free cash flow versus net profit guide explains why neither reported profit nor a single cash-flow subtotal answers every reinvestment question.

What this means for India

Indian businesses can participate through technology services, equipment, power, construction and operating infrastructure. But a global spending announcement is not a company-level revenue forecast.

The relevant connection is a disclosed role: a customer contract, project milestone, commissioned asset or recognised service. Then come payment terms, execution costs and funding needs. Our existing AI capex and India analysis follows that physical chain; this article addresses the separate economic question behind the bubble debate.

For IT services, faster delivery can lower the cost of work and also reduce the billable effort a customer buys. The AI productivity and Indian IT analysis examines who retains that saving.

The useful conclusion is a set of observable conditions

An improving case would combine paying demand, repeat business, viable unit economics and a funding plan that does not depend indefinitely on new outside capital. A weakening case might show demand that needs increasing subsidies, repeated delays, poorer collection or obligations growing faster than the cash available to meet them.

None of these observations, alone, proves a bubble. Nor does a falling share price prove that the technology failed. The point is to replace an argument about labels with a record of what would strengthen or challenge the business case.

Altys’s source-linked research and company monitoring can help revisit those observations as disclosures change. Keep the source, period and original assumption together, and use available Excel exports to check the analysis. Request access to Altys.

AI’s usefulness is one question. Who earns the cash, how long the assets remain useful and what investors assumed are three more. Serious research needs all four.

General financial education only. No securities recommendations, price targets, expected returns or portfolio allocations. Altys Labs is not a SEBI-registered Research Analyst or Investment Adviser.